Validate the type of target in binary samplers

[chkoar]

Some methods require that the type of target y is binary.
For these methods an option would be to use type_of_target from sklearn.utils.multiclass and raise a ValueError during fit if the target is not binary

[glemaitre]

@chkoar In fact we cannot do this. I started to implement it but the estimator in scikit-learn need to handle single-class target and multiclass target to pass the estimator test. Raising an error do not allow us to be compatible with scikit-learn base estimator.

However, I saw the exact same issue for polylearn and an issue is open in that regard. Let's keep this issue open until that the scikit-learn API get flexible.

[chkoar]

We could show a warning at least.

[glemaitre]

@chkoar Yep I agree with that. I already did something similar for single class target vector.

[chkoar]

Great

[glemaitre]

It will be link with the PR #112